The Fed Sought Early Access To Anthropic’s Mythos To Address Its Capabilities. It Didn’t Get It For Months

The Federal Reserve spent at least three months without access to Anthropic's Claude Mythos Preview, an advanced artificial intelligence model that federal officials had previously identified as a potential cybersecurity concern for the U.S. banking system.

The development comes amid heightened attention on cyber resilience across critical infrastructure following years of state-linked cyber incidents tied to conflicts involving Russia, Ukraine, and the Middle East, as well as increasing competition between the United States and China over advanced technologies.

Federal Reserve Chairman Kevin Warsh told lawmakers last week that the central bank had been seeking access to Mythos and other frontier AI models so it could identify and address potential vulnerabilities within its own systems and those connected to the broader financial sector, according to CNBC.

"We are not the deciders as to who has access, but I have not been shy in sharing my views with authorities across the government about the vulnerabilities," Warsh said during testimony before the Senate Banking, Housing and Urban Affairs Committee on July 15.

Warsh added that the Federal Reserve's concerns extend beyond a single model. "As these new models find their way more broadly, our banking system and frankly, the Federal Reserve, needs to do all we can to patch any vulnerabilities that we have," he told lawmakers, according to congressional testimony cited by CNBC.

Anthropic introduced Claude Mythos Preview and its associated cybersecurity initiative, Project Glasswing, in April. The company said the model was designed to identify weaknesses in software and cybersecurity defenses and initially made it available to a select group of approximately 50 organizations, including financial institutions and technology companies, according to Anthropic's Project Glasswing announcement.

Among the organizations publicly identified by Anthropic were JPMorgan Chase, Amazon, Apple, and Google. The company also said it had been in discussions with U.S. government entities, including the Cybersecurity and Infrastructure Security Agency (CISA) and the Center for AI Standards and Innovation, according to Anthropic.

The issue dates back to an extraordinary meeting convened in April by then-Federal Reserve Chair Jerome Powell and Treasury Secretary Scott Bessent with chief executives from major U.S. banks. Officials warned attendees that Mythos possessed capabilities that could expose software vulnerabilities at leading financial institutions.

Despite those concerns, the Federal Reserve itself did not have access to the model for months afterward. CNBC reported that the central bank was still attempting to secure access as recently as July 15. Neither the Fed nor Anthropic provided additional comment on whether access has since been granted.

The rollout of Mythos became more complicated in June when Anthropic temporarily suspended access to Mythos 5 and Fable 5, citing compliance with federal export control directives tied to national security authorities. Access was later restored to a limited number of trusted partners before restrictions were lifted entirely, according to The Information.

Daniel Newman, chief executive of research firm The Futurum Group, said he was surprised the Federal Reserve was not among the initial organizations granted access to Mythos.

"You would think that the financial institution that sort of drives all the policy for the rest of the financial institutions would be front and center of at least having a chance to evaluate the new technology," Newman told CNBC.

Anthropic expanded Project Glasswing in June, adding more than 150 organizations across 15 countries. The company has not publicly disclosed whether the Federal Reserve was included among those additional participants.

The Federal Reserve has not indicated when it first requested access to Mythos or whether it has subsequently received permission to use the model.